JPH01175325A - Method for encoding digital audio data - Google Patents

Abstract

PURPOSE:To attain a more excellent decoding by transmitting doubly a predictive parameter of especially high significance out of predictive parameters when digital audio data are data-compressed and transmitted. CONSTITUTION:The predictive parameter is obtained for every block of an A/D-converted sample and data Dt.G of a predictive residual, predictive coefficients k1-k3 and gain data G are regarded as the digital audio data in a current 8mm video and an encoding is executed at every one field period. At this time, the data of high significance are assigned to residual 93 words and the data of high significance are written doubly. When an error is generated in the data k1 and C of high significance, the data written doubly are substituted. Thus, even when an uncorrectable error is generated in the original predictive parameter, an audio signal without problems on auditory feeling can be reproduced and even when there is a limitation on a recording capacity, an error correcting capacity to the predictive parameter of high significance is improved and the excellent decoding can be attained.

Description

[Detailed description of the invention] The explanation will be given in the following order.

A. Field of industrial application B. Overview of the invention C. Prior art D. Problem to be solved by the invention E. Means for solving the problem (Fig. 1) (Figures 1 and 2) Effects of the Invention A Field of Industrial Application This invention relates to a method for encoding digital audio data.

B. Summary of the Invention This invention provides a method for compressing digital audio data and transmitting it, by doubly transmitting especially important quotient p-measurement parameters among prediction parameters, so that better decoding can be achieved. This is what I did.

C. Conventional technology For example, in 8 mm video, as an optional function, the audio signal is digitized into a PCM signal during recording, this PCM signal is recorded in the overscan section of the tape, and the reverse process is performed during playback. It is accepted that the original audio signal can be obtained by performing

In this case, if you increase the sampling frequency and the number of quantization bits of the PCM signal, you can record and reproduce audio signals with better characteristics, but if you do so, the number of bits to be recorded and reproduced will increase, making recording and reproduction impossible. It's gone.

Therefore, when recording, the number of bits of the PCM signal is compressed.
It has been proposed to expand the number of bits during playback so that excellent recording and playback characteristics can be obtained even if the number of bits on the tape is small.

AD is a method of bit compression/expansion.
There is a method called PCM.

FIG. 3 shows an example of a transmission system using ADPCM. In this example, each 64 consecutive samples of input data is treated as one block, and the prediction coefficients of the prediction filter are optimized for each block. This is the case when controlling the value. At this time, the main data bit-compressed for each sample of the input data is output, and auxiliary data related to the bit compression is output for each l block.

That is, in the same figure, (10) is the encoder, (3
0) is a signal transmission system, and (40) is a decoder. For example, if it is applied to PCMW 8mm video, the encoder (10) is installed in the recording string, and the decoder (40) is installed in the playback string. The transmission system (30) includes an error correction processing circuit, a two-rotation magnetic head, and the like.

Then, in the encoder (10), the digital data
14). In this case, the human data Xt is an analog audio signal that is linearly A/D! exchanged P
It is a CM signal, for example, the sampling frequency is 48k.
Hz, and the number of quantization bits is 16 bits. Further, as shown in FIG. 5, it is assumed that the data Xt is expressed as a fixed point with -1≦Xt<t and as a two's complement (the same applies to other data).

Further, the delay circuits (12) and (13) are for synchronizing the timing of the main data and the auxiliary data, and each has a delay time of one block period (
Therefore, strictly speaking, if the input value of the terminal (11) is Xt, the output of the delay circuit (13) will be Xt-128, but for the sake of complexity, it will simply be written as Xt).

Further, the prediction value x ren for the data Xt is extracted from the prediction filter (19), and this value 7t is sent to the subtraction circuit (14).
The subtraction circuit (14) extracts the difference between the value Xt and the difference 1) tt+t=xt−the difference. This value Dt is the error (prediction residual) of the predicted value t with respect to the input value X[. Therefore, the value D
Ideally, t is Dt = 0, which is also a small value for boat fishing, so even if the word length of the value t is, for example, 16 bits, as shown in FIG. 4, for example, Dt≧ When it is 0, all significant bits on the MSB side are “θ”
When L)L<0, all the bits become "l" and the remaining few bits on the LSB side become 0" or "l" corresponding to the difference between the value Xt and x. When the value Dt becomes a large value, the lower bits can be ignored.

Therefore, this value Dt is supplied to the gain control circuit (15) and multiplied by G (G≧1) to obtain a normalized value D
(・G, and this value G−Dt is the fresh quantization circuit (16)
and is requantized into, for example, 4 bits (1 bet5t-G).

Furthermore, this value et-c is supplied to the gain control circuit (17) and multiplied by 1/G, and is therefore made into an unnormalized value β (of the same order as the value Llt, and this value 5t
is supplied to the adder circuit (18), and the filter (
The predicted value from 19) is supplied to the adder circuit (18), and from the adder circuit ll8 (18), the value 5t and × original Japanese father L ×t==nu1+51 are taken out, and this value gt is supplied to the filter (19).

In this case, the value ×L is the predicted value for the value The value difference is approximately equal to the human power value Xt. Then, this value 5;1t is the filter (19
), the value F which is the output of the filter can be a value that predicts the input value Xt◆1 at the next sample time.

Then, the value e5t-G from the requantization circuit (16) is
The signal is supplied to a decoder (40) through a transmission system (30).

In this decoder (40), the value 5t-G is multiplied by 1/G by the gain control circuit (41) to obtain the value bt,
This value 5t is supplied to the adder circuit (42), and the added output is taken out to the output terminal (44), and is also supplied to the prediction filter (43) configured similarly to the filter (19), and the filter output is It is supplied to the adder circuit (42).

Therefore, the output of the filter (43) becomes the value xt, and at the same time, digital data approximately equal to the human input data Xt is taken out to the terminal (44).

Further, in order to set the prediction coefficients in the filters (19) and (43) to optimal values for each block, the following circuit is provided.

That is, the prediction filters (19) and (43) are
As a prediction coefficient, for example, partial autocorrelation coefficient (PARCO) l
The third-order filter uses coefficients (coefficients), and the first to third-order coefficients a1 to a3 can be changed to arbitrary values.

Further, the input data Xt from the terminal (11) is supplied to the time window circuit (21) to find the desired market, and then supplied to the autocorrelation circuit (22) to calculate the correlation coefficient. This coefficient is supplied to the prediction coefficient circuit (23), and a partial autocorrelation coefficient of 1 to 3 is calculated as a prediction coefficient up to the third order for each block of data Xt, and a value of 1 to 3 is calculated for this coefficient. It is supplied to the filter (19) and the latch (
51) to the filter (43).

Further, data Xt from the delay circuit (12) is supplied to a prediction error filter (24), and the filter output is supplied to an intra-block maximum value detection circuit (25).

In this case, the filter (24) is the prediction filter (19)
a third-order prediction filter (241) configured in the same manner as
a subtraction circuit (242), and a coefficient circuit (242).
The prediction coefficients 1 to 3 from 3) are supplied to the filter (241), and a predicted value (prediction error) f5t of the error L)t with respect to the input data Xt is generated for each sample. Further, the detection circuit (25) detects the prediction error i within each block of input data Xt.
5t (there are 64 of them), the absolute value 5max of the pre-1lJl+J error, which has the largest absolute value, is detected.

Then, this maximum value j5max is calculated by the normalized gain calculation circuit (
26) and the data of the gain G during normalization, Q x
b/j5max b is g<1. With a safety factor of <1, it is converted to, for example, b-0,9, and this data G is used in the gain control circuit (15).

(17) and is also supplied to the gain control circuit (41) through the latch (52). In this case, the value 1
Since this is the maximum value of 64 values 5t in 5IIIa, the value L)t-G is normalized to -1≦1)t-G<1.

Note that the latch (52) is 1 for data.
3, G for one block period of the corresponding value t5t-c;

Also, considering the amount of data transmitted from the encoder (10) to the decoder (40) through the transmission system (30), data 5L-G, which is the main data, is transmitted for each sample in 4 bits, and the auxiliary data The prediction coefficients 1 (1 to 3) and the gain data G, which are (prediction parameters), are transmitted for each block in, for example, 16 bits, 12 bits, 12 bits, and 8 bits, so the amount of data in one block period is 4 bits x 64 samples + 16 bits + 12 bits +
12 bits + 8 bits = 304 bits. and,
The amount of data for one block period when data is not compressed is 16 hints x 64 numbers - 1024 bits.

Therefore, the amount of data is 304 bits/1024 bits=29. '/% and then transmitted.

In this way, according to this system, data compression of digital audio data can be performed, but in this case,
In particular, according to this system, even if there is a limit on the double length of the coefficients and operations, the prediction coefficients of the prediction filters (19) and (43) are controlled to optimal values according to the input data XL, so that the decoded Errors caused by compression of data 5t can be minimized.

In addition, when transmitting the prediction residual Di, this residual 1)t is requantized to reduce the number of pits, and normalization is performed before the requantization, so that the transmitted data f5t- c is data with a small number of bits and a small number of errors.

FIGS. 5 and 6 show an example in which the encoder (10) and decoder (40) described above are used in a recording system and a reproducing system for 8 mm video audio signals.

However, in recording thread, for example, N 'I' S
A C color video signal is supplied to a recording video circuit (62) through a terminal (61), and a luminance signal of 1.
At the same time, the signal is converted twice, and the carrier frequency (c) is lower than the IQ sent false color signal and the M luminance signal, f c = 47.25 f h (# 743 kHz).
The frequency is converted to a signal with z, f h is the horizontal frequency),
A summed signal Sv of these l='M luminance signals, a low-frequency converted carrier color signal, and a biloft signal for tracking servo during playback is extracted, and this signal Sv is sent through a recording amplifier (63) to a switch circuit ( 64).

In addition, stereo left and right channel audio signals R and A/D converters (72L) and (7211) are connected through terminals (71L) and (71R).
＞, and each of H is A/D converted into digital data xt, xt with a sampling frequency of 48 kHz and a number of quantum bits of 16 bits 1-, for example.
These data xt, xt are sent to the encoder (10) mentioned above.
An encoder (IOL) configured similarly to (I
For each channel, data b(・
G, kt～3. G is taken out.

Then, these data are supplied to the recording encoder (73) and recorded for each field period, including addition of error correction data, interleaving, and time axis compression to 115 field periods at the end of each field period. The encoded digital signal Sa is supplied to a modulation circuit (74) and is converted into, for example, a pi-phase mark signal sb, and this signal sb is sent to a switch circuit (64) through a recording amplifier (75). is supplied to

Then, the switch circuit (64) is controlled at a predetermined timing so that the signal Sv is alternately supplied to the rotating magnetic heads (LA) and (1B) every field period, and the signal sb is different from the signal Sv. In the reverse relationship, the head (IA
), (IB).

Also, the heads (IA) and (IB) are 18
It has an angular interval of 0° and is rotated at a frame frequency in synchronization with the color video signal of the terminal (6■),
The magnetic tape (2) is run diagonally at a constant speed over an angular range of just over 216° around the U-axis circumference. In addition,
The heads (LA) and (IB) have different slit angles, so-called azimuth angles.

Therefore, as shown in FIG.
Tracks (2T) are successively formed adjacent to each other, and one track (2T) is formed in a 36° section from the beginning
The signal sb for the field period is recorded, and the remaining 180°
In this section, the signal Sv for one field period is recorded.

Note that the recording system and recording format described above are the same as those of the current 8 mm video except for the signal Sa in the signal sb.

On the other hand, in the reproduction system, the head (1^) sequentially reproduces the signals Sv and Sb from every other track (2T), and the remaining every other track (2T) is reproduced by the head (1^).
2T), the signals Sv and Sb are sequentially reproduced, and these reproduced signals are sent to the reproduction amplifier (81A').

(81B) to the switch circuit (82), and from the switch circuit (82) the head (IA),
The reproduced signals Sv, Sv of (IB) are taken out continuously, and the reproduced signals sb, sb of heads (IA), (IB) are taken out approximately every 115 field periods at the end of each field period. .

Then, the signal Sv from the switch circuit (82) is supplied to the reproduction video circuit (83), where the reverse processing to that during recording is performed and the original color video signal is taken out to the terminal (84).

Further, the signal sb from the switch circuit (82) is supplied to the demodulation circuit (91) to demodulate the signal Sa, and this signal Sa is supplied to the reproduction decoder (92) for time axis expansion and
By performing deinterleaving and error correction, the original data 15t-G,k is restored for each channel.
3. Cr is decoded, and these data are sent to a decoder (40) configured similarly to the decoder (40) described above.
L), (40R) is supplied to the data father t,5;
Ct is decoded, these data are supplied to the D/A converters (93L) and (93R), and the terminals (94
L), (94R) has the original audio signal,
R is taken out.

As described above, the audio signal R is recorded and reproduced. In this case, when calculating the amount of data in one field period, the amount of data of the prediction residual t5t-G is 48,000 samples per channel. Field frequency = 48000
/ 50.94...? The sample will be 800.8, but I can't round it down, so I'll set it to 80.
1 sample - 801X, t bits. Therefore, for 2 channels of signal R and H, 801 x 4 focus x 2 channels = 801 words (1
word = 8 bits).

Also, the amount of data for the prediction coefficients 1 to 3 and the gain data G is 1 set for every 64 samples (1 block), so for each channel, 801 samples/64
Samples = 12.51 sets, but since fractions cannot be rounded down, there are 13 sets, and for 2 channels of signal R and H, there are 26 sets. When this is converted into the number of words, it becomes (16+12+12+8 bits) x 26 sets = 6 words x 26 sets, 156 words.

Therefore, the total amount of data in one field period is 801 words + 156 words - 957 words.

On the other hand, when calculating the amount of digital audio data in current 8mm video, the sampling frequency is 2f h = 31.468kHz, fi
Since the number of quantization bits is 8 bits, the data density in the field period is 2 channels of signal and H, and 2 f h/field frequency x 2 channels = 525
Sample x 2 channels = 1050 words.

Therefore, since the amount of data by ADPCM mentioned above is smaller than the amount of PCM audio data in the current 8 mm video, the encoder (73) converts each data by the above ADPCM as it is into the digital audio data in the current 8 mm video. The recording can be encoded by assuming that Also, this allows the encoder (73) - tape (2) - decoder (9)
The signal thread of 2) is unchanged from the current 8mm video (that is, the current tape format can be used, but audio signals can be recorded and played back with better sound quality.

Literature: "Fundamentals of Speech Information Processing", patent application published by Ohmsha, 1986
-299285 Specification and Drawing D Problems to be Solved by the Invention By the way, in the above-mentioned 8 mm video, even if an error occurs in the demodulated signal Sa, the data 5t-G, If the error of 3 and G can be completely corrected in kt~, no problem will occur.

However, in reality, when performing encoding processing for recording in the encoder (73), only one set of P parity, Q parity, and CRC code is added for each field period for error correction. Therefore, complete error correction may not be possible with this method.

Then, 3. data 15t-G, kt~. As can be seen from the number of bits allocated to G, the prediction coefficient is 1
If there is an error in ~3, especially if there is an error in the low-order prediction coefficient 1 or gain data G,
The accuracy of the decoded data t is greatly reduced, and the auditory sense is greatly affected.

Therefore, data 1 to 3. It is necessary to add even more powerful error correction capability to G.

However, according to the above numerical example, the data capacity of the current 8 mm video in one field period is 1050 words, whereas the above At) PCM uses 957 words, so one field period uses 957 words. There is only 1050 words - 957 words = 93 words left over during the single word period, and within these 93 words, the data is 1 to 3. The error correction ability for G must be improved.

This invention attempts to answer such a need.

E Means for solving the problem Now, if we consider the partial autocorrelation coefficient of 1 to 3 as a prediction coefficient, its 6th coefficient kd is calculated by comparing certain data Xt with data Xt that is d samples apart. Therefore, in general, the second-order and third-order coefficients kz
, 1 is more important than 3 for the first-order coefficient. In other words, low-order coefficients have high importance.

Furthermore, since the gain data G indicates the gain G at the time of normalization, this is also highly important.

The present invention focuses on this point, and for example, all data t>t-c every l field period.

k1 ~ 3. G is encoded by regarding it as digital audio data in the current 8mm video, and at this time, data with high importance is assigned to the remaining 93 words, and the data with high importance is written in two wheels. It is something.

F When an error occurs in data kt or G with high action importance, the data written on the second wheel is substituted.

G Embodiment G1 First Embodiment In the example shown in FIG. 1, the encoder (73) extracts 1 for the prediction coefficient and gain data G for each block, and adds 1 to these data. G is inserted into the remaining area of the signal Sa.

In this case, each block of data kx, G is 24 bits, or 3 words, so in one field period, 3 words x 26 blocks = 78 words, and 1. G can be inserted into the remaining 93 word area.

The above encoding process is performed in the encoder ('/3). Next, the encoding process will be explained.

FIG. 1 shows the data allocation (memory map) in the memory used for the encoding process, and one address has a capacity of one word (8 pins). Also, the size of this memory is l field period, 1
It is 32 words x 12 words.

Then, at the time of encoding, i, prediction residual data 5t-c is 4 per sample.
Since the left and right channels form a pair, the left channel data 5L-G occupies the upper 4 bits, and the right channel data 5L-G occupies the lower 4 bits. The data are combined to form one piece of data Mi having a length of l words.

In this case, as described above, data f5t-c is obtained for 801 samples x 2 channels during the l-field period, and therefore data Mi is obtained for 801 samples during the l-field period.
Words will be obtained, so these 801 words can be converted into data Mo = Meoo (1 = 0 to 8
00).

110 Next, data Mi is sequentially written into the memory of FIG. 1 at the address locations shown in the figure.

iii. Also, as mentioned above, during the L field period, 6 words x 26 sets of data 1 to ka and G are obtained, so each word is converted into data Smn (m is data of 6 words per set). The word number m=0~5゜n is 2
Block numbers for 6 sets of data, n = Q ~ 25)
Then, this data Smn is sequentially written to address positions as shown in the figure.

iv, 1 to 3 to 26 sets of data. From G, the data is 1. as shown in FIG. G is taken out, and this is reconfigured into three words of data U1n, U2n+U3n for each set, and written to the memory as shown in the figure.

Note that in this case, actually, the writing process of terms ii to iv can be performed sequentially for each block.

v, 6-word user's data (ID word) IDo
= ID5 is written as shown in the figure.

For the data written in terms vi, ii to V, 132 pairs of P parity words Pj (j=0 to 131
) and Q parity words Qj are generated, and these parity words Pj, Qj are written to each address as shown.

CRC for the data of vti, ii = vi term
A code Cr (r=0 to 263) is generated and written to each address of the memory as shown in the figure. However, in this case, the CRC code Cr consists of two words, the even numbered code and the next odd numbered code, which is the original 1
It forms one CRC code.

viii. When the above writing is completed, the data in the memory shown in FIG. 1 is read out one word at a time in the column direction (vertical direction) from the data Qo in the first column.
Up to 32 columns are read out in the same way.

However, at the time of reading, a 3-bit marker MK and an 8-bit column address 80R3 are added to the beginning of each column of the read data, as shown in Figure 2. be done. In addition, in the same figure, data W is data lDoA-1l)51MiISllIn, U t
It is either n-U 3n.

Also, when reading this, time axis compression is performed by increasing the reading speed.Furthermore, since the data writing order and the reading order are different, interleaving is performed. The read data string is the signal Sa and is supplied to the modulation circuit (74).

The above-mentioned encoding process is for one field period, and in reality, it is necessary to process the audio signal R for more than one field period, so the memory shown in FIG. In these two memories, the above-mentioned encoding process is performed every one field period, shifted by one field period from each other, and the signals Sa are taken out one after another.

Note that in the above, data Mi + Smn, Ut
The parity word, CRC code, format of No. 1n Sa, etc. are the same as those of the current 8 mm video, except that n to U3n are data based on the predictive encoding method.

On the other hand, during decoding, the signal Sa from the 1,i9 tone circuit (91) is written to the same address in the memory (FIG. 1) in the same order as read during encoding. However, at this time, marker MK and address 80R5 in signal Sa are not written.

(2) Error data is detected using the 6P parity word Pj, and among the error flags (error pointers) prepared for each data, the error flag corresponding to the error data is sent.

■For data with error flags set,
Error correction is performed by P parity word Pj. When this error correction is correctly performed, the corresponding error flag is reset.

The same processing as in Items 1 and 11 is performed using the lV and Q parity words Qj.

■, When there is an error flag that is marked, ■
The processing in items ~■ is repeated. However, the maximum number of repetitions is, for example, five times.

Therefore, when this V term is finished, error data may exist, and the error flag of that error data is set.

Further, the processing up to this point is the same as that for current 8 mm video.

Vl, the error flag of the data Sen, for example, the error flag of the first prediction coefficient kl of the first block is checked and reset, since no error has occurred in that prediction coefficient. A 1 is read from the memory to the prediction coefficient, and a 1 to this coefficient is latched into the latch (51).

■When the error flag is set in the Vl term, the data U 1n −U 311 is written twice.
Among them, the error flag of the data corresponding to the data Smn, in this example, the error flag of the double-marked data UIO 1020 for the first prediction coefficient kl of the first block is checked and reset. Sometimes, the data U 101U20 is latched into a latch (51) as a prediction coefficient.

When the error flag is marked in section ■ or ■,
For example, a latch operation of 1 is not performed on the prediction coefficient for the latch (51), and therefore 1 is used as 1 for the prediction coefficient of the previous block as is for the prediction coefficient of the current block, that is, by holding the previous value. 1 is interpolated to the prediction coefficient.

IX, the processing of Vl to ■ is applied to all data of data Smn. The same is done for G. Also, the coefficient is 2. For example, a pre-hold is performed for an error of 3.

Then, the above processing is performed using two sets of memories as in the case of encoding, with a difference of one field period from each other for each field period, and the data t5t -G,
k t ~ 3. G is decoded, and based on this decoded data 5t-G, □~3°G, data R is
t, ¥: t is decoded and the signal R is taken out.

In this case, when an uncorrectable error occurs in the data Smn, the prediction coefficient is corrected or the gain data G is corrected by the data U 1n -U 311 by the processing in item (2), and the prediction coefficient is changed to 2. 3 is an interpolated value, but these coefficients k 2 r k 3 are less important than coefficient 1 and data G, so we decoded data father t using complete coefficients 1 to 3 and data G. Compared to the case, the error in the data 2t increases only slightly, and no problem arises in terms of audibility.

H Effects of the Invention Thus, according to the present invention, when digital audio data is recorded and reproduced by ADPCM, the prediction residual for each sample and the prediction parameter for each block are recorded, and the prediction parameters with high importance are recorded. Since the prediction parameters are recorded twice, even if an uncorrectable error occurs in the original prediction parameters, an audio signal with no audible problem can be reproduced.

In addition, instead of recording all of the prediction parameters on the two wheels, only the prediction parameters that are important are recorded in duplicate, so even if there is a limit to recording capacity, the importance Error correction ability for high prediction parameters is improved.

[Brief explanation of the drawing]

FIG. 1 is a memory map diagram of an example of the present invention, and FIGS. 2 to 7 are diagrams for explaining the same. (10) is an encoder, (30) is a signal transmission system, (40
) is a decoder.

Claims (1)

[Claims] A predetermined number of A/D-converted samples constitute one block, and for each block, a prediction parameter is determined from the samples included in that block, and based on this prediction parameter. Then, a prediction residual is calculated for each sample of the block for which this prediction parameter was calculated, and this prediction residual is sent out at the rate of the sample, and the prediction parameter is sent out at the rate of the block. , A method for encoding digital audio data, in which a prediction parameter with a high degree of importance among the prediction parameters is transmitted twice in the proportion of the blocks.